Product Management Case Study

We taught Duolingo learners to actually speak.

This is the story of how our team designed the Duolingo Peer Platform, a feature that helps learners cross the gap between a long streak and a real conversation. It started with one uncomfortable truth we kept hearing in interviews: people felt good using the app, but they still froze the moment they had to talk.

Role: Product Manager, on a team of six
Product Management, Carnegie Mellon University, Silicon Valley
Six week project, six month product roadmap
Day 27 streak "Let's talk" Duolingo characters taking a group photo together
0
learners interviewed
0
pain points scored
0
personas we built for
0
final grade out of 10
Chapter 01 · The Brief

A simple assignment with a hard question hiding inside it

Build a real product concept for Duolingo, from customer research all the way to a launch roadmap. We picked the company because everyone on the team already used it. That turned out to be the interesting part.

Our team, the Birdkeepers, was made up of six people who all had a green owl living somewhere on their phone. Some of us had streaks in the hundreds. And yet, when we sat down and talked honestly, almost none of us felt like we could hold a conversation in the language we were supposedly learning.

That gap between how much we used the app and how little we could actually say became the whole project. Instead of inventing a flashy new feature and hoping people wanted it, we decided to start from the frustration we already felt and go find out whether other learners felt it too.

So we set a plan. We would talk to real students, sort their frustrations by hand, score them with a framework so our opinions did not run the show, and only then design something. Everything you read below happened in that order, and this page walks through it the same way we lived it.

The one sentence version

Learners build strong habits on Duolingo but plateau before they can hold a real conversation. We designed a peer to peer speaking feature, guided by AI, to close that gap.

Chapter 02 · The Problem

A long streak is not the same thing as fluency

Duolingo is brilliant at building a daily habit. What it does not do, at least not yet, is put you in front of another human and let you stumble through a real exchange until it clicks.

The phrase we heard again and again, in slightly different words every time, was that the app is fun but limited for real conversations. People loved the streaks, the little animations, the sense of showing up. But when a friend asked them to say something in Spanish or French, they went quiet.

The more we listened, the more the problem sorted itself into four honest complaints. None of them are secrets. All of them are the reason so many learners quietly drift away after the beginner stage.

  • Habits without depth. A streak proves you opened the app. It does not prove you can order a coffee abroad without freezing.
  • Practice that feels like a textbook. Tapping the right tiles is satisfying, but it rarely feels like the messy back and forth of a real chat.
  • Real practice is expensive. Human tutors cost real money and need scheduling, which most students do not have to spare.
  • Nobody is keeping you accountable. Learning alone is easy to quit, and most people quit right after the honeymoon phase.

This matters because it is not a small niche. Roughly one in three learners drops out after the beginner stage, and the biggest reason they give is the lack of real conversation practice. That is a lot of people who wanted to learn, built a habit, and left anyway.

A Duolingo character studying with a mentor, papers flying with checkmarks
Plenty of effort and gold stars. Still stuck when the conversation is real.

"Even learners with long streaks admitted they did not feel fluent. That is the gap between consistency and confidence."

A pattern that showed up in almost every interview
Chapter 03 · Talking to Learners

We stopped guessing and started listening

Before we let ourselves design anything, we ran interviews with real students across several universities and programs. The goal was not to confirm what we already believed. It was to find out where we were wrong.

We wrote a discussion guide of open ended questions, careful not to lead people toward the answers we wanted. Each question was tied to a hypothesis about why students learn, what makes them stop, and whether the app was actually building real ability or just a comforting routine.

To keep up with the volume, we used Fireflies to capture transcripts and ChatGPT to help us condense messy notes into themes. We treated the AI as a fast intern, not an oracle, and cleaned up every transcript by hand before we trusted a single quote.

0

Interviews

Real students, one conversation at a time.

0

Universities

Different campuses, different pressures.

0

Programs

From computer science to studio art.

0

Locations

Spread across the country.

Two people in an interview setting, one taking notes
Every insight on this page traces back to a real conversation like this one.

What one interview taught us

One learner, working toward a government recognized French diploma, said something that reframed the whole project. She had tried Duolingo, ignored the notifications, and eventually deleted it. What finally made her serious was paying for structured classes with real people.

Her point was not that Duolingo was bad. It was that habit alone did not make her show up, and it did not give her the explanations, the structure, or the human practice she needed to feel ready. Money and people made her accountable in a way a push notification never could.

"Money is a better motivator than notifications. Once I paid and had real classes, I actually showed up."

Kushi, French learner, interviewed for this study
Chapter 04 · The People We Met

Four learners we could not stop thinking about

We started with the lazy label of "college students" and quickly threw it out. The real opportunity was a sharper group, learners with strong habits but low confidence, and two subgroups inside it who would try peer practice first.

Alex Ramirez

Mainstream learner · 24 · New York

"I want to reconnect with my heritage and actually use Spanish when I travel."

A marketing professional who squeezes five minute lessons into his morning coffee and evening commute. He needs practice that is affordable, adaptive, and easy to keep up with a busy life. The free tier is what got him in the door.

SpanishBusy scheduleHabit builder

Emily Chen

International grad student · 24 · Los Angeles

"Language is the first barrier to making friends and working in this country."

A computer science masters student from China. She reads and writes English confidently, but seminars, career fairs, and casual small talk make her anxious. She wants practice that fits between coursework and job applications, and that builds real speaking confidence.

EnglishAcademicCareer prep

Jake Miller

Sophomore · 20 · New York University

"I want to turn my commute into real learning instead of doomscrolling."

Diagnosed with ADHD and easily pulled away by short form video. He starts lessons on the subway and gets interrupted constantly. He needs short, novel tasks that match his attention span, plus real accountability so a broken streak does not end the whole thing.

NeurodiverseBite sizedNeeds accountability

Maya Thompson

Comparative literature · 19 · On campus

"I learn languages for K-pop lyrics, anime, and Spanish films."

A fandom driven learner who finds generic vocabulary boring. She wants niche words tied to the culture she loves, and she wants to share progress with her online community. For her, learning is part of her identity, not a chore.

Korean, Japanese, SpanishFandomSocial

Once we had these four in front of us, the target got obvious. We narrowed the launch to university students who already had the habit but not the confidence, with international and neurodiverse learners as the first people we would win over.

Chapter 05 · Making Sense of the Noise

From twelve complaints down to two that mattered most

Every learner had opinions, and every opinion felt urgent in the moment. To keep our own bias out of it, we scored all twelve pain points with the RICE framework, then filtered them again with MoSCoW.

RICE forces you to be honest. For each pain point we asked how many people it reaches, how much impact solving it would have, how confident we were, and how much effort it would take. The math does not care about your favorite idea, which is exactly the point.

Pain pointReachImpactConfidenceEffortRICE
Shallow learning impact 554425
Weak conversational practice 454420
Ad heavy free tier535325
Notification fatigue524220
Gamification pressure424216
High subscription cost434412
Limited advanced learning support344412
Unclear long term value443412
Exercise redundancy424311
Minimal native speaker influence34349
Lack of formal recognition23357
Unnecessary social features31426

A few things scored high on raw RICE, including the ad heavy free tier. But when we ran everything through MoSCoW and weighed real learner value against business impact, two pain points rose to the top as genuine must haves. They also happened to be two sides of the same coin.

Must have

  • Shallow learning impact
  • Weak conversational practice

Should have

  • Ad heavy free tier
  • Notification fatigue

Could have

  • Gamification pressure
  • High subscription cost

Won't have yet

  • Unnecessary social features
  • Formal recognition

The verdict

Shallow learning impact and weak conversational practice. Solve real speaking, and you quietly solve the depth problem too.

Chapter 06 · The Idea

The Duolingo Peer Platform

A university focused space where learners join short, guided speaking events with real peers, and an AI coach rides along to keep the conversation flowing and hand back honest feedback at the end.

We considered three directions. One leaned into contextual conversation practice. Another focused on student friendly pricing and certificates. A third built calmer, ADHD friendly engagement. All three were good, but only one hit both of our must have pain points at once, and hit them in a way no competitor was doing.

So we committed to conversational and contextual learning, delivered as a peer platform. The magic is the blend. Real humans bring the unpredictability and warmth of an actual conversation. The AI brings structure, gentle prompts when things stall, and feedback that would normally cost a tutor.

Two Duolingo characters having a lively conversation with speech bubbles
The whole point: get learners talking to each other, safely and often.

Peer matching

Learners are paired by proficiency, goals, and context, whether that is studying abroad or prepping for a career fair, for quick sessions of ten to twenty minutes.

Fifteen minute guided sessions

An AI guide sets the scene and offers prompts so the conversation never dies, from ordering food to a mock interview to campus small talk.

Real time AI feedback

Instant, private notes on pronunciation, fluency, and clarity, with a clean summary the moment the session ends.

Just enough gamification

Streaks, points, and conversation milestones to keep learners coming back, without the pressure that made people resent the old notifications.

The core loop

Browse events, join one, practice with a peer, receive AI feedback, then come back for the next. Simple enough to build as an MVP, sticky enough to become a habit.

Chapter 07 · Try It Yourself

The app, running right here

Talking about a speaking feature only goes so far. So here is a working prototype inside a phone. Tap through the flow, or use the steps on the left, and watch a learner go from browsing events to getting real feedback.

Follow the core loop

This is the exact journey we designed for. Each screen is interactive, so click around the same way a learner would on a Tuesday night before a trip.

This concept was also built as a separate clickable prototype during the course. You can open that version too.

Open the v0 prototype
9:41
Peer Practice
27
English to Spanish · Level B1 · Near you
Café small talk
15 min · 3 spots left
Tap to join
Networking practice
15 min · starts in 4 min
Set reminder
Travel: ordering food
15 min · beginner friendly
Tap to join

Café small talk

Order a drink, make small talk, react naturally

15 minute guided session
Matched with 1 peer at your level
AI prompts if the chat goes quiet
Private feedback, only you see it
Join now
Back to events
Live session · 06:12
You
Sofía
AI guide

Try asking Sofía what she usually orders and why. Then react to her answer.

End session

Nice work

Here is your private session summary

Fluency+12%
PronunciationStrong
ClarityGood
Great pace and confidence. Keep an eye on gender agreement, you said "el mesa" instead of "la mesa".
Back to events
Practice
Events
Progress
Chapter 08 · The Business Case

Why this is worth building, not just nice to have

A good feature still has to earn its place. The Peer Platform turns Duolingo from a beginner friendly app into a full learning ecosystem, and it feeds the same flywheel that already makes the company work.

Duolingo already runs on two loops. More learners create more data, which sharpens the product, which brings in more learners. More engagement drives more paid subscribers, which funds more investment. Speaking practice pours fuel on both, because it is exactly the kind of high value habit that keeps people around and nudges them toward Super and Max.

Diagram of Duolingo's learning flywheel and investment flywheel meeting at more learners
Our feature strengthens both the learning flywheel and the investment flywheel at the same time.

The market backs it up. Language learning is on its way from roughly 21 billion dollars to 44 billion by 2030, pushed by AI and community driven learning. Here is how we sized our slice of it.

TAM
1.8B

Language learners worldwide

SAM
200M

Mobile first learners across the major apps

SOM
25M

College aged Duolingo users, starting in the US

North star metric

Weekly speaking minutes per learner. It is the cleanest signal that the feature is working, and it maps directly to retention and upgrades.

How it pays off

Free learners get limited sessions and basic feedback. Super and Max unlock premium matching, themed events, and richer AI coaching, which lifts revenue per user.

Chapter 09 · Sizing Up the Field

Everyone solves part of this. Nobody solves all of it

We looked hard at who else was helping people speak. Each option nails one piece and drops another. The gap in the middle is exactly where the Peer Platform lives.

HelloTalk
Direct competitor
Unmoderated and no structured AI feedback
We add safe, guided sessions with real coaching
Rosetta Stone
Direct competitor
Expensive tutoring that does not scale for students
We keep peer practice affordable and flexible
YouTube
Indirect competitor
Free and rich, but completely passive
We make it interactive with measurable progress
Duolingo Video Call
In house alternative
Solo AI practice, episodic and Max only
We add real humans and real accountability

"We combine cultural authenticity, community accountability, and structured feedback at scale. That mix is what nobody else offers."

Our competitive thesis, in one line
Chapter 10 · The Roadmap

Crawl, then walk, then run

We did not try to ship everything at once. The plan proves value cheaply first, grows engagement second, and only scales to market leadership once the hard parts, like safety and privacy, are solid.

Crawl
0 to 2 months · test value and fit
  • Short 10 to 15 minute events for daily scenarios
  • AI feedback on pronunciation and fluency basics
  • English to Spanish, US university students
Plus 15% weekly speaking practice
Walk
2 to 4 months · expand and engage
  • Longer 20 to 30 minute events across contexts
  • Richer, personalized feedback and tips
  • More language pairs, working professionals
Plus 20 to 30% fluency, better retention
Run
4 to 6 months · integrate and lead
  • Personalized event recommendations
  • In session AI guidance and structured prompts
  • All languages, full US base over 16
Plus 20% daily active users
Chapter 11 · How It Landed

The work held up

We delivered a full product plan, a clickable prototype, and a final readout. The feedback we got back told us the story hung together, from the research all the way to the roadmap.

9.25/10

Average across the written plan and the final presentation

4.75
CSAT
45+
interviews
"Nice alignment from market to users to MVP. Great RICE analysis in detail. Small changes, high impact, with great slide design and a strong Crawl, Walk, Run roadmap."
Professor feedback, Product Management
"Loved the team photo on the cover, very creative. Really enjoyed the live demo at the end."
Teaching team
Chapter 12 · What I Took Away

The lessons that outlasted the grade

Some of these are about product. Some are about how a team of six actually gets to a decision. All of them stuck with me.

Community is a catalyst

Social practice made gamification and retention stronger. It was never an add on, it was the point.

Prioritization is the real skill

Prioritization is not just for features. It is how a team manages its own flood of ideas without stalling.

Watch the small players

Studying scrappy startups like HelloTalk early surfaced opportunities the big names never showed us.

Sequence the MVP

Every feature felt essential. Learning to ship by user value, in order, was harder and more useful than any single idea.

If I ran this project again, I would spend more time sharpening our research objectives before touching any tools, and I would set milestone tracking earlier so the last week did not carry so much weight. But I would keep the core instinct that made it work. We started from a frustration we genuinely felt, we let real learners correct us, and we let a framework, not the loudest voice, pick the winner.

Want to see it move?

Play with the prototype above, or open the separate clickable version we built during the course.

Open the v0 prototype